Virtual network mapping for cloud services under probabilistic regional failures

Network virtualization is critical for distributing cloud services over expanded distances and improving scalability and responsiveness. However many providers are very concerned about maintaining high availability, particularly under large catastrophic/disaster type conditions causing multiple failures. Now recent studies have looked at network virtualization for disaster recovery. However these efforts have only treated resource-intensive protection strategies and have not incorporated the probabilistic nature of disasters. Hence this paper studies virtual network embedding using a-priori failure state information and proposes various strategies based upon both traffic engineering (resource efficiency) and risk minimization (disaster mitigation) objectives. The performance of these schemes is tested using network simulation and future research directions identified.